ENHANCING SHIP DETECTION IN FOGGY MARITIME ENVIRONMENTS: FOG DATA SYNTHESIS VS. DEFOGGING

In the event of a maritime distress accident, aircraft are used to efficiently search for ships over a wide area. However, fog often occurs in the ocean, which reduces visibility, making it difficult for aircraft to detect ships. To solve this problem, this research goal is to improve the performance of a ship detection model that can operate reliably in foggy environments. To this end, two approaches were compared: developing a ship detection model by training a deep learning-based object detection model on foggy data, and removing fog from input images using a defogging algorithm before detection. As a result, the Foggy model, trained using fog data, showed overall superior performance under real foggy conditions compared to the Sunny model, trained only on clear days. In particular, in environments with heavy fog, Recall and AP50 improved by 0.303 and 0.314, respectively. On the other hand, when the defogging algorithm was applied to remove fog from input images, detection performance improved in synthetic fog environments but deteriorated in real fog conditions due to color and contrast distortions. These results suggest that training with foggy data is effective for stable ship detection in real maritime environments and demonstrate that models trained in foggy conditions can improve detection performance in actual scenarios. This study is expected to improve the effectiveness of ship detection systems during maritime searches, even in foggy conditions.